Chemistry-informed recommender system to predict optimal molecular receptors in SERS nanosensors.

Multiple molecular receptors amplify signal variance from various receptor-analyte interactions, enhancing the specificity of surface-enhanced Raman scattering (SERS) detection. Currently, the number and type of receptors are manually determined based on chemical intuition and trial-and-error experimentation, often leading to inefficiency, errors, and missed opportunities. Here, we design a chemistry-informed SERS receptor recommender system (RS) powered by a three-stage 'identify, rank, and recommend' XGBoost framework. Our RS predicts the best receptors to interact with structurally similar haloanisoles and generate distinct SERS superprofiles for enhanced differentiation, attaining over 95% accuracy. Leveraging collaborative filtering with our RS database, we further showcase receptor recommendations for an unidentified haloanisole based on its molecular structure and chemical reactivity, even before experimentally collecting its SERS data. This systematic and data-driven methodology based on cheminformatics represents a shift from empirical discovery to an efficient approach for receptor optimization for complex sensing applications involving structurally similar analytes.